Introduction
On June 23, 2025 [2] [3] [4], a landmark ruling by US District Court Judge William Alsup in the case of Bartz v. Anthropic PBC addressed the application of fair use in the context of training large language models (LLMs) with copyrighted works [3]. This decision has significant implications for the legal landscape surrounding generative AI systems [6], particularly concerning the use of copyrighted materials [5] [6].
Description
On June 23, 2025 [2] [3] [4], US District Court Judge William Alsup issued a significant ruling in Bartz v [2]. Anthropic PBC [1] [2] [3] [4] [5] [7] [8], addressing the application of fair use in the context of training large language models (LLMs) with copyrighted works [3]. The court determined that while Anthropic’s initial downloading of over seven million books from piracy websites, including those in the LibGen database, constituted copyright infringement [2] [4], the subsequent use of those materials to train its generative AI model, Claude [1] [7] [8], was deemed “exceedingly transformative” and aligned with the fair use doctrine under Section 107 of the Copyright Act. The ruling marks a pivotal development in the legal landscape surrounding generative AI systems [6], emphasizing that the creation of a permanent library from pirated materials does not justify such actions, and Anthropic may face legal consequences for these unauthorized downloads.
The ruling allowed Anthropic to file a pre-class certification motion focused on fair use [3], facilitating an early resolution of key legal questions [3]. The court characterized the use of copyrighted works for training as transformative [3], noting that the LLMs did not reproduce or distribute the texts but instead learned language patterns [3]. Internal communications revealed that Anthropic’s leadership was aware of the illegality of their actions, opting to “steal” books to circumvent legal complications—a strategy not unique to Anthropic but observed in other AI companies as well.
Despite the ruling on transformative use [6], the court determined that Anthropic must face trial regarding its liability for downloading millions of pirated books [6], raising questions about the legality of its data acquisition methods [6]. A separate trial is scheduled for December to assess damages related to the pirated materials, which could reach up to $150,000 per work for willful infringement [7]. The court recognized the potential for a licensing market for AI training but clarified that copyright holders cannot monopolize all transformative uses of their works [3]. Judge Alsup’s experience with fair use cases [5], particularly from his involvement in Google v [5]. Oracle [5], positions this ruling as potentially influential for ongoing and future AI copyright lawsuits [5].
However, the court did not address whether downloading from a pirate site could be considered fair use for training purposes [8], nor did it resolve whether copies made from Anthropic’s central library [9], which were not used for training [9], constituted fair use [3] [4] [7] [8] [9], citing insufficient record development due to the company’s inadequate cooperation during the discovery process [9]. This ruling highlights ongoing concerns among copyright owners, who contend that AI companies are unlawfully reproducing their works [7], threatening their livelihoods [7]. The case was brought forth by authors Andrea Bartz [1], Charles Graeber [1] [4] [7], and Kirk Wallace Johnson [1] [4] [7], who accused Anthropic of large-scale copyright infringement [1], claiming the company profited from the unauthorized use of their creative works [1].
Anthropic expressed satisfaction with the ruling on transformative use but disagreed with the judge’s stance on the downloading of books and is considering its options for appeal [6]. The court noted that Anthropic’s later purchase of a book it had previously pirated would not absolve it of liability but could influence the assessment of statutory damages [8]. Further legal proceedings regarding other uses of library copies remain open [8], and an appeal is anticipated, with expectations that the findings regarding fair use may be overturned [9], as they appear to contradict established case law [9]. For AI developers [3], the ruling emphasizes the importance of lawful data acquisition and output filtering to avoid infringement [3], while copyright owners should focus enforcement efforts on acquisition and output, distinguishing between public-facing and internal uses [3]. This case is part of a broader trend of litigation concerning the use of copyrighted material in AI training [6], with other cases highlighting similar issues, underscoring the complex legal framework for determining fair use.
Conclusion
The ruling in Bartz v. Anthropic PBC underscores the evolving legal challenges faced by AI developers and copyright holders in the realm of generative AI. It highlights the necessity for AI companies to adopt lawful data acquisition practices and for copyright owners to strategically enforce their rights. As the legal landscape continues to develop, this case may serve as a precedent for future disputes, influencing how fair use is interpreted in the context of AI training.
References
[1] https://www.cbsnews.com/news/anthropic-ai-copyright-case-claude/
[2] https://www.transparencycoalition.ai/news/judge-hands-down-major-ruling-in-anthropics-ai-copyright-case-read-the-full-order-here
[3] https://fisherbroyles.com/news/client-alert-fair-use-copyright-and-ai-training-key-insights-from-the-anthropic-decision/
[4] https://news.bloomberglaw.com/ip-law/ai-training-is-fair-use-judge-rules-in-anthropic-copyright-suit-38
[5] https://www.wired.com/story/anthropic-ai-copyright-fair-use-piracy-ruling/
[6] https://deadline.com/2025/06/anthropic-fair-use-copyright-1236442323/
[7] https://tech.yahoo.com/ai/articles/anthropic-wins-key-ruling-ai-133440779.html
[8] https://barrysookman.com/2025/06/24/anthropic-ai-decision-on-copyright-fair-use/
[9] https://authorsguild.org/news/mixed-decision-in-anthropic-ai-case/
